Add local spoken passages with Google retrieval and isolated runtime setup

tools/test_package_windows.py

@@ -10,6 +10,7 @@ import zipfile
 from package_windows import package, CRT_LIBRARIES
 from embeddinggemma2 import runtime_provenance as provenance
 import stage_extraction_worker as extraction
+import stage_speech_worker as speech
 
 
 def executable(path, machine=0x8664):
@@ -32,6 +33,29 @@ def inventory(resources):
 
 
 class PackageTests(unittest.TestCase):
+    def test_speech_cpu_only_missing_and_tampered_runtime_preserve_previous_package(self):
+        for mutation in ('cpu', 'missing', 'tampered', 'stale', 'unlisted'):
+            with self.subTest(mutation=mutation), tempfile.TemporaryDirectory() as directory:
+                root, resources, app, output = self.fixture(directory)
+                output.write_bytes(b'previous package')
+                speech_directory = resources / 'speech-runtime'
+                if mutation == 'cpu':
+                    path = speech_directory / 'runtime-manifest.json'
+                    data = json.loads(path.read_text())
+                    data.update(device='cpu', distribution=False)
+                    path.write_text(json.dumps(data))
+                elif mutation == 'missing':
+                    (speech_directory / 'searchi-speech-worker.exe').unlink()
+                elif mutation == 'tampered':
+                    (speech_directory / 'ggml-vulkan.dll').write_bytes(b'changed')
+                elif mutation == 'stale':
+                    (root / 'tools/speech-runtime/worker.cpp').write_bytes(b'changed source')
+                else:
+                    executable(speech_directory / 'obsolete-ggml-cpu.dll')
+                with self.assertRaises(ValueError):
+                    package(root, app, output)
+                self.assertEqual(output.read_bytes(), b'previous package')
+
     def test_extraction_missing_tampered_stale_preserves_existing_package(self):
         for mutation in ("missing", "tampered", "stale"):
             with self.subTest(mutation=mutation), tempfile.TemporaryDirectory() as directory:
@@ -119,6 +143,28 @@ class PackageTests(unittest.TestCase):
         helper = root / "build/searchi-extract-worker.exe"
         executable(helper)
         extraction.stage(helper, extraction.fingerprint(root), resources, root)
+        speech_directory = resources / 'speech-runtime'
+        speech_directory.mkdir()
+        for name in speech.INPUTS:
+            path = speech.input_path(root, name)
+            path.parent.mkdir(parents=True, exist_ok=True)
+            path.write_bytes(speech.input_path(speech.ROOT, name).read_bytes())
+        for name in speech.NOTICES:
+            original = root / 'assets/licenses' / name
+            original.parent.mkdir(parents=True, exist_ok=True)
+            original.write_bytes((speech.ROOT / 'assets/licenses' / name).read_bytes())
+            (speech_directory / name).write_bytes(original.read_bytes())
+        speech_files = []
+        for name in ('searchi-speech-worker.exe', 'ggml-cpu-x64.dll', 'ggml-vulkan.dll',
+                     'whisper.dll', 'ggml.dll', 'ggml-base.dll', 'vulkan-1.dll',
+                     'VULKAN-LICENSE.txt', 'MSVC-REDISTRIBUTION.txt', *sorted(CRT_LIBRARIES)):
+            path = speech_directory / name
+            executable(path)
+            speech_files.append({'path': name, 'bytes': path.stat().st_size, 'sha256': speech.digest(path)})
+        (speech_directory / 'runtime-manifest.json').write_text(json.dumps({
+            'schema': 1, 'upstream_commit': speech.UPSTREAM, 'device': 'auto', 'distribution': True,
+            'source_hashes': {name: speech.digest(speech.input_path(root, name)) for name in speech.INPUTS},
+            'files': speech_files}))
         return root, resources, app, root / "archive.zip"
 
     def test_valid_pe_media_mutation_and_notice_corruption_preserve_previous_archive(self):